libevocosm::evoreal Class Reference

Tools for evolving real numbers. More...

#include <evoreal.h>

Inheritance diagram for libevocosm::evoreal:

List of all members.

Public Member Functions
	evoreal (float a_sign_weight=5.0F, float a_exponent_weight=5.0F, float a_mantissa_weight=90.0F)
	Creation constructor.
	evoreal (evoreal &a_source)
	Copy constructor.
evoreal &	operator= (evoreal &a_source)
	Assignment.
float	mutate (float a_f)
	Mutation for float values.
double	mutate (double a_d)
	Mutation for double values.
float	crossover (float a_f1, float a_f2)
	Crossover for float values.
double	crossover (double a_d1, double a_d2)
	Crossover for double values.

Additional Inherited Members
Static Protected Member Functions inherited from libevocosm::globals
static void	set_seed (uint32_t a_seed)
	Set the seed for the random number generator.
static uint32_t	get_seed ()
	Set the seed for the random number generator.
static std::string	version ()
	Get version number.
static size_t	rand_index (size_t n)
	Static function to allow use of g_random function pointer in random_shuffle.

---

Detailed Description

The majority of genetic algorithms work on pure bit strings, converting those strings to the desired types for fitness testing. In Lawrence Davis' book "Handbook of Genetic Algorithms," he transforms a 44-bit string into two floating point values via a series of operations. I've seen similar techniques elsewhere, and I find them a bit cumbersome.

In the purest sense, a GA should have no knowledge of the format of the data it is modifying; however, natural chromosomes do encode some structure in their sequence; for example, crossover appears to take place in specific positions along the chromosome. And while mutation doesn't care about a chromosome's structure, it does affect that structure. In context of a computer program, the structure of a chromosome isn't so important as the ability to logically modify its bits through crossover and mutation.

I built tools for the mutation and crossover of encoded floating-point values of types float and double. The code that follows assumes we are working with 32-bit floats and 64-bit IEEE-754 doubles, which, in my experience, the norm for many C and C++ compilers. Yes, I'm aware of the VAX and other systems; this code is explicitly non-portable outside implementations of IEC 60559/IEEE-754.

---

Constructor & Destructor Documentation

libevocosm::evoreal::evoreal	(	float	a_sign_weight = 5.0F,
		float	a_exponent_weight = 5.0F,
		float	a_mantissa_weight = 90.0F
	)

Creates a new evoreal object based on a set of weights that define the chance of mutation in various components of a floating-point value. The default weights have worked well in a variety of applications, but are (of course) settable for specific application and experimentation.

Each weight is a percentage of the total of all three weights; for example, if the three weights add to 100 (as they do by efault), and a_sign_weight is 12, the chance of a mutation in the sign bit is 12%. The default weights were chosen based on experience in using these tools in a variety of applications.

Parameters:

a_sign_weight	- Weight assigned to changes in sign
a_exponent_weight	- Weight assigned to changes in the exponent
a_mantissa_weight	- Weight assigned to changes in the mantissa

libevocosm::evoreal::evoreal

(

evoreal &

a_source

)

Creates a new evoreal with the same states as an existing one.

Parameters:

a_source

- The source object

---

Member Function Documentation

float libevocosm::evoreal::crossover	(	float	a_f1,
		float	a_f2
	)

Creates a new float by combining two values through a real-specialized form of crossover.

Parameters:

a_f1	- First parent number
a_f2	- Second parent number

Returns:

A combination of the two arguments

double libevocosm::evoreal::crossover	(	double	a_d1,
		double	a_d2
	)

Creates a new double by combining two values through a real-specialized form of crossover.

Parameters:

a_d1	- First parent number
a_d2	- Second parent number

Returns:

A combination of the two arguments

float libevocosm::evoreal::mutate

(

float

a_f

)

Returns a new float that is a mutated version of the argument.

Parameters:

a_f	- Number to be cloned; the result is then mutated

Returns:

A clone of a_f that has bene mutated

double libevocosm::evoreal::mutate

(

double

a_d

)

Returns a new float that is a mutated version of the argument.

Parameters:

a_d	- Number to be cloned; the result is then mutated

Returns:

A clone of a_d that has bene mutated

evoreal& libevocosm::evoreal::operator=

(

evoreal &

a_source

)

Assigns the state of one evoreal to another.

Parameters:

a_source

- The source object

---

The documentation for this class was generated from the following file:

• evoreal.h